INTERNATIONAL CENTER FOR EQUITY IN HEALTH UNIVERSIDADE FEDERAL DE PELOTAS Wednesday, 19 June Practice 3: Concentration index and Slope index of inequalities Duration: 2h In this practical session we aim to explore means of quantifying the degree of income-related inequality in a specific health variable, focusing in two indicators: the concentration index and the slope index of inequality. Concentration index (CIX) The CIX is defined with reference to the concentration curve, using an approach analogous to the Gini coefficient. It graphs on the x-axis the cumulative percentage of the sample, ranking individuals according to socioeconomic position. The cumulative percentage of the health variable is plotted on the y-axis. 1. Using Microsoft Excel ® Before using the Stata codes, we will compute the CIX and its standard error (SE) from grouped data. For this purpose, we are going to use an Excel spreadsheet available among the files previously provided. Please, make sure that you have the file ‘concentration_index.xls’. The concentration index is easily computed in a spreadsheet program using the following formula1: C = (p1L2-p2L1) + (p2L3-p3L2) + … + (pT-1LT-pTLT-1) Where p is the cumulative percent of the sample ranked by economic status – we will use the wealth index quintiles as indicator of economic status; L(p) is the corresponding concentration curve ordinate; and T is the number of socioeconomic groups. The file has three sheets. In the first one, we have the question: ‘Do you have standard deviations for the group means?’. It will help you decide which one to use. If you do not have the SD of the estimates, use the sheet 'no std devs'; if you do have it, use sheet 'have std devs'. We will estimate the CIX for ‘stunting’ (height-for-age < -2SD), using the wealth quintiles dataset (wiq DHS + MICS). For Albania (MICS 2005), as an example, we have the following estimates: country year source wiqn stunt_r stunt_se stunt_N stunt_pop Albania 2005 MICS Q1 0.351764 0.031577 214 243.9873 Albania 2005 MICS Q2 0.280018 0.036474 211 231.5362 Albania 2005 MICS Q3 0.283129 0.036225 193 192.7294 Albania 2005 MICS Q4 0.219511 0.030918 239 228.683 Albania 2005 MICS Q5 0.181064 0.02602 222 183.4465 1 N Kakwani, A Wagstaff & E van Doorslaer "Socioeconomic inequalities in health: measurement, computation and statistical inference", Journal of Econometrics 77 (1997) 87-103. Page 1 of 5 INTERNATIONAL CENTER FOR EQUITY IN HEALTH UNIVERSIDADE FEDERAL DE PELOTAS We will use the sheet ‘no std devs’, in which you will find the box below. ENTER DATA BELOW # persons quintile Quintile per quintile means 1 2 3 4 5 wiq stunt_pop stunt_r We have to enter the group sizes and estimates. No other data is needed. Note that the other cells are locked and they will be calculated automatically based on the data entered. If you select a locked cell, you will be able to visualize the hidden formula in the toolbar. We also highlight that we use the expanded population equivalent instead of the absolute sample size, as it already considers the sample design through the adjustment for the sample weights. 1. Select the variables stunt_pop and stunt_r for Albania (MICS 2005) in your datasheet. 2. Copy the values and paste in the sheet ‘no std devs’ of the file ‘concentration_index.xls’. The column ‘# persons per quintile’ is equals to the stunt_pop in our dataset, as the ‘quintile means’ is equal to the stunt_r column. 3. Take note of the value of concentration index estimated. What does it mean? 2. Using Stata codes We can estimate the CIX using the Stata command cixr, which uses the following syntax: cixr <socioeconomic position> <health variable> [if] [in], [graph] Where the wealth index quintiles are used as indicators of socioeconomic position; and health variable indicates any intervention or outcome. You can also include conditions using the [if] option and/or draw a concentration curve using the [graph] option. Let’s estimate the CIX of stunting for Albania (MICS 2005), similarly to what we did using Excel. In this case, we will write the syntax as follows: First we need to take into account that variable wiq is in a string format and we need to generate a numeric one: sort country year wiq by country year: gen wiqn= _n Then, type cixr wiqn stunt_r in 1/5, graph Note: You must have the ado file cixr in the folder ‘c:\ado\personal’ of your computer. The variable wiqn must be ordinal and ordered so that low is poor/worst and high is rich/best. We use the option ‘in 1/5’ to Page 2 of 5 INTERNATIONAL CENTER FOR EQUITY IN HEALTH UNIVERSIDADE FEDERAL DE PELOTAS indicate that we want to include only the lines from 1 to 5, in which we have the data for Albania (MICS 2005). 4. Run the cixr command using the syntax provided above. 5. Did you obtain the same estimate as using the Excel spreadsheet? 6. What does the graph show? 7. Now we aim to compare estimates of CIX for the same indicator across different countries/surveys. Estimate the cix of stunting prevalence for the Belarus (MICS 2005) and Vietnam (MICS 2010). How do you compare these estimates? What would you say about the income-related inequality in stunting among these countries? Slope index of inequalities (SII) The SII is estimated through a different approach. It is typically derived through linear regression, but for prevalence indicators, it is more appropriate to use logistic regression. The independent variable in this regression are the midpoints of ranks obtained by ordering the sample by the explanatory variable when using grouped data. The ranks are scaled so that the values range from zero to one. Therefore, the SII represents the absolute difference in the fitted value of the health indicator between the top (score of 1) and the bottom (score of 0) of the socioeconomic scale. We can estimate the SII using the Stata command siilogit, which uses the following syntax: siilogit <socioeconomic position> <health variable> [if] [in], [nograph] where the wealth quintiles are used as indicators of socioeconomic position; and health variable indicates any intervention or outcome. You can also include conditions using the [if] option and the command generates a graph that allows checking the linearity as a default. You can use the option [nograph] if the graph is not necessary. Let’s estimate the SII of stunting for Albania (MICS 2005). In this case, we will write the command as follows: siilogit wiqn stunt_r in 1/5 Note: You must have the ado file siilogit in the folder ‘c:\ado\personal’ of your computer. The variable wiqn must be ordinal and ordered so that low is poor/worst and high is rich/best. We use the option ‘in 1/5’ to indicate that we want to include only the lines 1 to 5 from the dataset, in which we have the data for Albania (MICS 2005). Page 3 of 5 INTERNATIONAL CENTER FOR EQUITY IN HEALTH UNIVERSIDADE FEDERAL DE PELOTAS 8. Run the siilogit command using the syntax provided above. 9. How would you interpret the SII? 10. What does the graph show? 11. Now we aim to compare estimates of SII for the same indicator across different countries/surveys. Estimate the SII of stunting prevalence for the Belarus (MICS 2005) and Vietnam (MICS 2010). How do you compare these estimates? What would you say about the income-related inequality in stunting among these countries? Now lets estimate both CIX and SII for different indicators for the same country. 12. Estimate CIX and SII for measles vaccination and underweight using data from the Nigeria (DHS 2008). 13. Analyze the estimates and graphs. How would you interpret them? What would you say about the income-related inequality in both indicators? Page 4 of 5 INTERNATIONAL CENTER FOR EQUITY IN HEALTH UNIVERSIDADE FEDERAL DE PELOTAS Templetes .8 0 .2 .4 .6 Cumulative percent .6 .4 .2 0 0 .2 .4 .6 .8 Cumulative percent ranked by economic status 1 0 .2 .4 .6 .8 Cumulative percent ranked by economic status 1 0 .2 .4 .6 .8 1 Vietnam (MICS 2010) Cumulative percent Cumulative percent .8 1 Belarus (MICS 2005) 1 Albania (MICS 2005) 0 .2 .4 .6 .8 Cumulative percent ranked by economic status 1 Page 5 of 5